[PATCH] ALSA: hda/hdmi: Disable silent stream on GLK
From: Ville Syrjälä ville.syrjala@linux.intel.com
The silent stream stuff recurses back into i915 audio component .get_power() from the .pin_eld_notify() hook. On GLK this will deadlock as i915 may already be holding the relevant modeset locks during .pin_eld_notify() and the GLK audio vs. CDCLK workaround will try to grab the same locks from .get_power().
Until someone comes up with a better fix just disable the silent stream support on GLK.
Cc: stable@vger.kernel.org Cc: Harsha Priya harshapriya.n@intel.com Cc: Emmanuel Jillela emmanuel.jillela@intel.com Cc: Kai Vehmanen kai.vehmanen@linux.intel.com Cc: Takashi Iwai tiwai@suse.de Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/2623 Fixes: 951894cf30f4 ("ALSA: hda/hdmi: Add Intel silent stream support") Signed-off-by: Ville Syrjälä ville.syrjala@linux.intel.com --- sound/pci/hda/patch_hdmi.c | 21 +++++++++++++++------ 1 file changed, 15 insertions(+), 6 deletions(-)
diff --git a/sound/pci/hda/patch_hdmi.c b/sound/pci/hda/patch_hdmi.c index 415701bd10ac..ffcde7409d2a 100644 --- a/sound/pci/hda/patch_hdmi.c +++ b/sound/pci/hda/patch_hdmi.c @@ -2947,7 +2947,8 @@ static int parse_intel_hdmi(struct hda_codec *codec)
/* Intel Haswell and onwards; audio component with eld notifier */ static int intel_hsw_common_init(struct hda_codec *codec, hda_nid_t vendor_nid, - const int *port_map, int port_num, int dev_num) + const int *port_map, int port_num, int dev_num, + bool send_silent_stream) { struct hdmi_spec *spec; int err; @@ -2980,7 +2981,7 @@ static int intel_hsw_common_init(struct hda_codec *codec, hda_nid_t vendor_nid, * Enable silent stream feature, if it is enabled via * module param or Kconfig option */ - if (enable_silent_stream) + if (send_silent_stream) spec->send_silent_stream = true;
return parse_intel_hdmi(codec); @@ -2988,12 +2989,18 @@ static int intel_hsw_common_init(struct hda_codec *codec, hda_nid_t vendor_nid,
static int patch_i915_hsw_hdmi(struct hda_codec *codec) { - return intel_hsw_common_init(codec, 0x08, NULL, 0, 3); + return intel_hsw_common_init(codec, 0x08, NULL, 0, 3, + enable_silent_stream); }
static int patch_i915_glk_hdmi(struct hda_codec *codec) { - return intel_hsw_common_init(codec, 0x0b, NULL, 0, 3); + /* + * Silent stream calls audio component .get_power() from + * .pin_eld_notify(). On GLK this will deadlock in i915 due + * to the audio vs. CDCLK workaround. + */ + return intel_hsw_common_init(codec, 0x0b, NULL, 0, 3, false); }
static int patch_i915_icl_hdmi(struct hda_codec *codec) @@ -3004,7 +3011,8 @@ static int patch_i915_icl_hdmi(struct hda_codec *codec) */ static const int map[] = {0x0, 0x4, 0x6, 0x8, 0xa, 0xb};
- return intel_hsw_common_init(codec, 0x02, map, ARRAY_SIZE(map), 3); + return intel_hsw_common_init(codec, 0x02, map, ARRAY_SIZE(map), 3, + enable_silent_stream); }
static int patch_i915_tgl_hdmi(struct hda_codec *codec) @@ -3016,7 +3024,8 @@ static int patch_i915_tgl_hdmi(struct hda_codec *codec) static const int map[] = {0x4, 0x6, 0x8, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf}; int ret;
- ret = intel_hsw_common_init(codec, 0x02, map, ARRAY_SIZE(map), 4); + ret = intel_hsw_common_init(codec, 0x02, map, ARRAY_SIZE(map), 4, + enable_silent_stream); if (!ret) { struct hdmi_spec *spec = codec->spec;
Hi,
On Wed, 22 Dec 2021, Ville Syrjala wrote:
The silent stream stuff recurses back into i915 audio component .get_power() from the .pin_eld_notify() hook. On GLK this will deadlock as i915 may already be holding the relevant modeset locks during .pin_eld_notify() and the GLK audio vs. CDCLK workaround will try to grab the same locks from .get_power().
Until someone comes up with a better fix just disable the silent stream support on GLK.
decoupling the call to get_power() is a non-trivial change (especially as it's done from generic hda_codec.c code), so I'd say let's go with this patch for GLK:
Reviewed-by: Kai Vehmanen kai.vehmanen@linux.intel.com
Br, Kai
On Wed, 22 Dec 2021 15:53:50 +0100, Ville Syrjala wrote:
From: Ville Syrjälä ville.syrjala@linux.intel.com
The silent stream stuff recurses back into i915 audio component .get_power() from the .pin_eld_notify() hook. On GLK this will deadlock as i915 may already be holding the relevant modeset locks during .pin_eld_notify() and the GLK audio vs. CDCLK workaround will try to grab the same locks from .get_power().
Until someone comes up with a better fix just disable the silent stream support on GLK.
Cc: stable@vger.kernel.org Cc: Harsha Priya harshapriya.n@intel.com Cc: Emmanuel Jillela emmanuel.jillela@intel.com Cc: Kai Vehmanen kai.vehmanen@linux.intel.com Cc: Takashi Iwai tiwai@suse.de Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/2623 Fixes: 951894cf30f4 ("ALSA: hda/hdmi: Add Intel silent stream support") Signed-off-by: Ville Syrjälä ville.syrjala@linux.intel.com
Thanks, applied now.
Takashi
participants (3)
-
Kai Vehmanen
-
Takashi Iwai
-
Ville Syrjala